Assessing reproducibility of metabolomic profiles generated using a multi-lab round-robin" study design"

Authors

• HA Chatelaine
• C Beecher
• EA Mathé

Abstract

Establishing reproducibility in metabolomics results is required to ensure reliable clinical conclusions, both by accurately identifying metabolites and reporting reproducible relative abundances. The TruQuant platform is a potential solution to address this challenge and enhance clinical utility of metabolomics approaches. This study assesses the reproducibility of the TruQuant approach when multiple labs, using their own LC-MS parameters, collect TruQuant data on the same set of samples. Eight labs received identical aliquots from 32 samples, containing multiple known mixtures of beef, pork, and chicken extracts, each mixture analyzed in duplicate, from which they collected metabolomics data using their own methods. Extract mixtures serve as a case-study for metabolically heterogenous samples, as would be encountered in clinical settings. Data was analyzed for: 1) data quality, 2) reproducibility across labs, and 3) prediction of sample types from metabolite profiles. Metabolite abundances between duplicate samples within each lab had high correlation values (all but one with r > 0.9) and low total drift in metabolite abundance over run time, ensuring high quality data. Despite differences in numbers of features detected, metabolomic profiles from each of the 8 labs reproducibly differentiated sample mixture types via unsupervised clustering analyses (principal component analysis and unsupervised random forest). Machine learning prediction models of data mixtures from each lab are ongoing. Overall, our “round-robin” study design and our data show that the TruQuant approach enables reproducible measurements across different labs and platforms, a first step to enabling reproducible clinical application of metabolomics approaches.

Scientific Focus Area: Computational Biology

This page was last updated on Tuesday, August 6, 2024